Vacuum and Large Objects

On 12/15/2011 04:01 AM, Simon Windsor wrote:

Hi

I am having problems recovering storage from a Postgres 9.05 database
that is used to hold large XML blocks for a week, before they are
archived off line.

The main tables are partitioned in daily partitions, and these are
easy to manage, however the DB keeps growing despite using Vacuum
(daily at 0700) and autovacuum (this does not seem to run, although
the process is running). The system is insert only, and partitions are
dropped when over 7 days of age.

I believe the issue lies with pg_largeobject, it is split between 88
files of approx. 1G each.

Are you using pg_largeobject via the lo_ functions, or via the `lo'
datatype?

If you're using it via the `lo' type then certain actions can IIRC cause
large object leaks. Try vacuumlo.

http://www.postgresql.org/docs/current/static/vacuumlo.html

vacuumlo is **NOT** suitable for use on databases where you use the lo_
functions directly.

See also the `lo' module:

http://www.postgresql.org/docs/current/static/lo.html

If you're using the lo_ functions directly and still seeing excessive
space consumption in pg_largeobject then beyond a CLUSTER or VACUUM FULL
run I'm not sure what to advise.

--
Craig Ringer

From: Simon Windsor [mailto:simon.windsor@cornfield.me.uk]
Sent: Wednesday, December 14, 2011 3:02 PM
To: pgsql-general@postgresql.org
Subject: Vacuum and Large Objects

Hi

I am having problems recovering storage from a Postgres 9.05 database that is used to hold large XML blocks for a week, before they are archived off line.

The main tables are partitioned in daily partitions, and these are easy to manage, however the DB keeps growing despite using Vacuum (daily at 0700) and autovacuum (this does not seem to run, although the process is running). The system is insert only, and partitions are dropped when over 7 days of age.

I believe the issue lies with pg_largeobject, it is split between 88 files of approx. 1G each.

The Postgres settings are default, EXCEPT

grep ^[a-z] postgresql.conf
listen_addresses = '*'          # what IP address(es) to listen on;
port = 5432                             # (change requires restart)
max_connections = 1000                  # (change requires restart)
shared_buffers = 256MB                  # min 128kB
work_mem = 4MB                          # min 64kB
maintenance_work_mem = 256MB            # min 1MB
vacuum_cost_delay = 20ms                # 0-100 milliseconds
checkpoint_segments = 32                # in logfile segments, min 1, 16MB each
checkpoint_completion_target = 0.9      # checkpoint target duration, 0.0 - 1.0
checkpoint_warning = 60s                # 0 disables
archive_mode = off              # allows archiving to be done
constraint_exclusion = partition        # on, off, or partition
log_destination = 'stderr'              # Valid values are combinations of
logging_collector = on          # Enable capturing of stderr and csvlog
silent_mode = on                        # Run server silently.
log_checkpoints = on
log_line_prefix = '%t %d %u '                   # special values:
log_statement = 'none'                  # none, ddl, mod, all
track_activities = on
track_counts = on
autovacuum = on                 # Enable autovacuum subprocess?  'on'
log_autovacuum_min_duration = 250       # -1 disables, 0 logs all actions and
autovacuum_max_workers = 3              # max number of autovacuum subprocesses
autovacuum_naptime = 3min               # time between autovacuum runs
autovacuum_vacuum_threshold = 500       # min number of row updates before
autovacuum_analyze_threshold = 100      # min number of row updates before
autovacuum_vacuum_scale_factor = 0.1    # fraction of table size before vacuum
autovacuum_analyze_scale_factor = 0.05  # fraction of table size before analyze
autovacuum_vacuum_cost_delay = 5ms      # default vacuum cost delay for
autovacuum_vacuum_cost_limit = 200      # default vacuum cost limit for
statement_timeout = 0                   # in milliseconds, 0 is disabled
datestyle = 'iso, dmy'
lc_messages = 'en_GB.UTF-8'                     # locale for system error message
lc_monetary = 'en_GB.UTF-8'                     # locale for monetary formatting
lc_numeric = 'en_GB.UTF-8'                      # locale for number formatting
lc_time = 'en_GB.UTF-8'                         # locale for time formatting
default_text_search_config = 'pg_catalog.english'

Besides running VACUUM FULL  pg_largeobject;, is there a way I can get autovacuum to start and clear this up?

All the best

Simon

Simon Windsor
Eml: simon.windsor@cornfield.org.uk
Tel: 01454 617689
Mob: 07590 324560

"There is nothing in the world that some man cannot make a little worse and sell a little cheaper, and he who considers price only is that man's lawful prey."

I might be a bit late in this discussion.
But I think, your problem is right here:

" running VACUUM FULL pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

Regards,
Igor Neyman

Hi Igor
2011/12/16 Igor Neyman <ineyman@perceptron.com> wrote: > But I think,
your problem is right here:

" running VACUUM FULL  pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

I'm pretty sure that VACUUM FULL builds new indexes. That's at least
of how I understand the docs, especially the first "tip" here
http://www.postgresql.org/docs/9.0/static/routine-vacuuming.html

Yours, Stefan

2011/12/16 Igor Neyman <ineyman@perceptron.com>:

On Fri, 2012-01-06 at 07:12 +0100, Stefan Keller wrote:

Hi Igor
2011/12/16 Igor Neyman <ineyman@perceptron.com> wrote: > But I think,
your problem is right here:

" running VACUUM FULL pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

I'm pretty sure that VACUUM FULL builds new indexes. That's at least
of how I understand the docs, especially the first "tip" here
http://www.postgresql.org/docs/9.0/static/routine-vacuuming.html

Before 9.0, VACUUM FULL required a REINDEX afterwards if you want to
keep decent performances.

With 9.0, it is no longer required because the new VACUUM FULL doesn't
bloat the index anymore.

So, in a sense, you were both right :) The documentation you're
referring to is the 9.0 manual. And Igor specified that one need to
REINDEX after VACUUM FULL for any release prior to 9.0. Both right.

--
Guillaume
http://blog.guillaume.lelarge.info
http://www.dalibo.com
PostgreSQL Sessions #3: http://www.postgresql-sessions.org

Stefan,

-----Original Message-----
From: Stefan Keller [mailto:sfkeller@gmail.com]
Sent: Friday, January 06, 2012 1:12 AM
To: Igor Neyman
Cc: Simon Windsor; pgsql-general@postgresql.org
Subject: Re: Vacuum and Large Objects

Hi Igor
2011/12/16 Igor Neyman <ineyman@perceptron.com> wrote: > But I think,
your problem is right here:

" running VACUUM FULL  pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow

it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

I'm pretty sure that VACUUM FULL builds new indexes. That's at least of
how I understand the docs, especially the first "tip" here
http://www.postgresql.org/docs/9.0/static/routine-vacuuming.html

Yours, Stefan

2011/12/16 Igor Neyman <ineyman@perceptron.com>:

From: Simon Windsor [mailto:simon.windsor@cornfield.me.uk]
Sent: Wednesday, December 14, 2011 3:02 PM
To: pgsql-general@postgresql.org
Subject: Vacuum and Large Objects

Hi

I am having problems recovering storage from a Postgres 9.05 database

that is used to hold large XML blocks for a week, before they are
archived off line.

The main tables are partitioned in daily partitions, and these are

easy to manage, however the DB keeps growing despite using Vacuum
(daily at 0700) and autovacuum (this does not seem to run, although the
process is running). The system is insert only, and partitions are
dropped when over 7 days of age.

I believe the issue lies with pg_largeobject, it is split between 88

files of approx. 1G each.

The Postgres settings are default, EXCEPT

grep ^[a-z] postgresql.conf
listen_addresses = '*'          # what IP address(es) to listen on;
port = 5432                             # (change requires restart)
max_connections = 1000                  # (change requires restart)
shared_buffers = 256MB                  # min 128kB work_mem =

4MB

# min 64kB maintenance_work_mem = 256MB            # min 1MB
vacuum_cost_delay = 20ms                # 0-100 milliseconds
checkpoint_segments = 32                # in logfile segments, min 1,
16MB each checkpoint_completion_target = 0.9      # checkpoint target
duration, 0.0 - 1.0 checkpoint_warning = 60s                # 0
disables archive_mode = off              # allows archiving to be

done

constraint_exclusion = partition        # on, off, or partition
log_destination = 'stderr'              # Valid values are
combinations of logging_collector = on          # Enable capturing of
stderr and csvlog silent_mode = on                        # Run

server silently.

log_checkpoints = on
log_line_prefix = '%t %d %u '                   # special values:
log_statement = 'none'                  # none, ddl, mod, all
track_activities = on track_counts = on autovacuum =

on

# Enable autovacuum subprocess?  'on'
log_autovacuum_min_duration = 250       # -1 disables, 0 logs all
actions and autovacuum_max_workers = 3              # max number of
autovacuum subprocesses autovacuum_naptime = 3min               #

time

between autovacuum runs autovacuum_vacuum_threshold = 500       # min
number of row updates before autovacuum_analyze_threshold = 100

#

min number of row updates before autovacuum_vacuum_scale_factor =

0.1

# fraction of table size before vacuum

autovacuum_analyze_scale_factor

= 0.05  # fraction of table size before analyze
autovacuum_vacuum_cost_delay = 5ms      # default vacuum cost delay
for autovacuum_vacuum_cost_limit = 200      # default vacuum cost
limit for statement_timeout = 0                   # in milliseconds,

0 is disabled datestyle = 'iso, dmy'

lc_messages = 'en_GB.UTF-8'                     # locale for system
error message lc_monetary = 'en_GB.UTF-8'                     #

locale

for monetary formatting lc_numeric = 'en_GB.UTF-

8'

# locale for number formatting lc_time = 'en_GB.UTF-

8'

# locale for time formatting default_text_search_config =

'pg_catalog.english'

Besides running VACUUM FULL  pg_largeobject;, is there a way I can

get autovacuum to start and clear this up?

All the best

Simon

Simon Windsor
Eml: simon.windsor@cornfield.org.uk
Tel: 01454 617689
Mob: 07590 324560

"There is nothing in the world that some man cannot make a little

worse and sell a little cheaper, and he who considers price only is
that man's lawful prey."

I might be a bit late in this discussion.
But I think, your problem is right here:

" running VACUUM FULL  pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow

it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

Regards,
Igor Neyman

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To
make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

I qualified my reply with PG version specifics, since original poster didn't specify Postgres version he's using.

Regards,
Igor Neyman

-----Original Message-----
From: Stefan Keller [mailto:sfkeller@gmail.com]
Sent: Friday, January 06, 2012 1:12 AM
To: Igor Neyman
Cc: Simon Windsor; pgsql-general@postgresql.org
Subject: Re: Vacuum and Large Objects

Hi Igor
2011/12/16 Igor Neyman <ineyman@perceptron.com> wrote: > But I think,
your problem is right here:

" running VACUUM FULL  pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow

it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

I'm pretty sure that VACUUM FULL builds new indexes. That's at least of
how I understand the docs, especially the first "tip" here
http://www.postgresql.org/docs/9.0/static/routine-vacuuming.html

Yours, Stefan

2011/12/16 Igor Neyman <ineyman@perceptron.com>:

From: Simon Windsor [mailto:simon.windsor@cornfield.me.uk]
Sent: Wednesday, December 14, 2011 3:02 PM
To: pgsql-general@postgresql.org
Subject: Vacuum and Large Objects

Hi

I am having problems recovering storage from a Postgres 9.05 database

that is used to hold large XML blocks for a week, before they are
archived off line.

The main tables are partitioned in daily partitions, and these are

easy to manage, however the DB keeps growing despite using Vacuum
(daily at 0700) and autovacuum (this does not seem to run, although the
process is running). The system is insert only, and partitions are
dropped when over 7 days of age.

I believe the issue lies with pg_largeobject, it is split between 88

files of approx. 1G each.

The Postgres settings are default, EXCEPT

grep ^[a-z] postgresql.conf
listen_addresses = '*'          # what IP address(es) to listen on;
port = 5432                             # (change requires restart)
max_connections = 1000                  # (change requires restart)
shared_buffers = 256MB                  # min 128kB work_mem =

4MB

# min 64kB maintenance_work_mem = 256MB            # min 1MB
vacuum_cost_delay = 20ms                # 0-100 milliseconds
checkpoint_segments = 32                # in logfile segments, min 1,
16MB each checkpoint_completion_target = 0.9      # checkpoint target
duration, 0.0 - 1.0 checkpoint_warning = 60s                # 0
disables archive_mode = off              # allows archiving to be

done

constraint_exclusion = partition        # on, off, or partition
log_destination = 'stderr'              # Valid values are
combinations of logging_collector = on          # Enable capturing of
stderr and csvlog silent_mode = on                        # Run

server silently.

log_checkpoints = on
log_line_prefix = '%t %d %u '                   # special values:
log_statement = 'none'                  # none, ddl, mod, all
track_activities = on track_counts = on autovacuum =

on

# Enable autovacuum subprocess?  'on'
log_autovacuum_min_duration = 250       # -1 disables, 0 logs all
actions and autovacuum_max_workers = 3              # max number of
autovacuum subprocesses autovacuum_naptime = 3min               #

time

between autovacuum runs autovacuum_vacuum_threshold = 500       # min
number of row updates before autovacuum_analyze_threshold = 100

#

min number of row updates before autovacuum_vacuum_scale_factor =

0.1

# fraction of table size before vacuum

autovacuum_analyze_scale_factor

= 0.05  # fraction of table size before analyze
autovacuum_vacuum_cost_delay = 5ms      # default vacuum cost delay
for autovacuum_vacuum_cost_limit = 200      # default vacuum cost
limit for statement_timeout = 0                   # in milliseconds,

0 is disabled datestyle = 'iso, dmy'

lc_messages = 'en_GB.UTF-8'                     # locale for system
error message lc_monetary = 'en_GB.UTF-8'                     #

locale

for monetary formatting lc_numeric = 'en_GB.UTF-

8'

# locale for number formatting lc_time = 'en_GB.UTF-

8'

# locale for time formatting default_text_search_config =

'pg_catalog.english'

Besides running VACUUM FULL  pg_largeobject;, is there a way I can

get autovacuum to start and clear this up?

All the best

Simon

Simon Windsor
Eml: simon.windsor@cornfield.org.uk
Tel: 01454 617689
Mob: 07590 324560

"There is nothing in the world that some man cannot make a little

worse and sell a little cheaper, and he who considers price only is
that man's lawful prey."

I might be a bit late in this discussion.
But I think, your problem is right here:

" running VACUUM FULL  pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should follow

it with the "REINDEX TABLE ...", at least on PG versions prior to 9.0.

Regards,
Igor Neyman

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To
make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

Please, disregard my previous message.
I missed " Postgres 9.05 database" in original post.

Regards,
Igor Neyman

On Fri, 06 Jan 2012 08:51:24 +0100, Guillaume Lelarge wrote:

On Fri, 2012-01-06 at 07:12 +0100, Stefan Keller wrote:

Hi Igor
2011/12/16 Igor Neyman <ineyman@perceptron.com> wrote: > But I
think,
your problem is right here:

" running VACUUM FULL pg_largeobject"

If you are running "VACUUM FULL ..." on the table, you should

follow it with the "REINDEX TABLE ...", at least on PG versions prior
to 9.0.

I'm pretty sure that VACUUM FULL builds new indexes. That's at least
of how I understand the docs, especially the first "tip" here
http://www.postgresql.org/docs/9.0/static/routine-vacuuming.html

Before 9.0, VACUUM FULL required a REINDEX afterwards if you want to
keep decent performances.

With 9.0, it is no longer required because the new VACUUM FULL
doesn't
bloat the index anymore.

So, in a sense, you were both right :) The documentation you're
referring to is the 9.0 manual. And Igor specified that one need to
REINDEX after VACUUM FULL for any release prior to 9.0. Both right.

--
Guillaume
http://blog.guillaume.lelarge.info
http://www.dalibo.com
PostgreSQL Sessions #3: http://www.postgresql-sessions.org

Not quite true. I have develop server "PostgreSQL 9.0.5, compiled by
Visual C++ build 1500, 64-bit", after clearing LO and vacumming
accessing db stats from pgadmin gives permission denied on rel xxxx and
in fact this file gets wrong permissions. REINDEX resolves problem.

Regards,
Radek